Voice AI answers a law firm’s phones 24/7 for $200 to $500 a month, qualifying callers and booking consultations at less than a tenth of what a receptionist costs. The catch is discipline: disclose on every call, and never let it give legal advice.
Sometime in the past 18 months, AI phone voices crossed a line. Not “impressive for a robot.” Indistinguishable. Callers at my firms regularly get through an entire qualification conversation without realizing no human was on the line. A few have asked the AI for her extension so they could call her back directly.
Once callers can’t tell, the economics of every law firm phone line change. The phone has always been the most expensive part of intake and the least reliable: staffed 40 hours a week, dark the other 128, and dependent on whoever answers being sharp at 4:55 PM on a Friday. Voice AI removes the staffing constraint entirely. Here’s what that looks like in practice, because I run it every day.
What voice AI handles today
A properly configured voice agent answers on the first ring, every time, at 2 PM or 2 AM. No hold music, no voicemail, no “our office is currently closed.”
On the call itself, it qualifies the caller against your criteria: practice area fit, jurisdiction, timeline, whatever your intake sheet asks. It books consultations directly onto your calendar. It answers the questions that eat receptionist hours, like what you charge, how the process works, and how long cases take. When something urgent or sensitive comes through, it routes the call to a human immediately instead of taking a message.
And it logs everything. Every call lands in your CRM with a full transcript, a summary, and the structured data your intake team needs. No sticky notes, no “I think someone called about a lemon law thing yesterday.” If your phone platform and CRM don’t connect natively, Zapier closes that gap.
That last part matters more than the answering. A receptionist’s memory of a call is a paraphrase. A transcript is evidence of exactly what the caller said and exactly what your firm told them.
The platforms
Three ways in, depending on how much you want to build.
Purpose-built legal answering. Smith.ai’s AI receptionist is the fastest path for a firm that wants nothing to do with configuration. It comes trained on law firm call patterns, integrates with Clio and the usual CRMs, and a human team backs up the AI. You pay for the convenience, but setup is days, not weeks.
Contact-center platforms with AI agents. Aloware is what my firms use for calling infrastructure. Platforms in this class give you the phone system and the AI in one place: routing, SMS, power dialing for your human team, and AI agents on inbound. Right fit for firms doing real call volume that want intake, follow-up, and outbound in one system.
Roll-your-own. ElevenLabs conversational agents, Vapi, and Retell let you build exactly the agent you want: your script, your voice, your routing logic. This is the cheapest option at scale and the most flexible, but somebody technical has to own it. If you have that person, or you’ve reached stage five of the roadmap and can build with Claude Code yourself, it’s the strongest option.
| Platform | Type | Setup | Cost profile |
|---|---|---|---|
| Smith.ai | Legal-specific AI receptionist | Days, no build | Highest, zero technical work |
| Aloware | Contact center plus AI agents | Weeks, some config | Mid, scales with call volume |
| ElevenLabs / Vapi / Retell | Build your own | Needs a technical owner | Cheapest at scale |
How my firms run it
My lemon law firm signs clients in all 50 states, and the intake operation is AI-first on the phones. Overflow and after-hours calls never hit voicemail. A caller at 11 PM on a Sunday gets answered, qualified, and booked, and my intake team walks in Monday to a queue of scheduled consultations with transcripts attached instead of a voicemail box of half-audible callbacks.
My TCPA practice went further. We built our own AI call-handling tools with escalation routing by scenario: the AI works the routine calls end to end, and specific trigger situations route to a named human on my team, not a general queue. The AI doesn’t decide whether something is important. The scenario rules do, and we wrote them.
Neither firm fired anyone to do this. The humans moved up the chain, which is the point of the next two sections.
The disclosure question
Some states now regulate this directly. California’s bot disclosure law requires telling people when they’re talking to AI in certain commercial contexts, and its newer AI rules push the same direction. More states will follow.
My position: disclose on every call, in every state, whether or not a statute makes you. “You’ve reached Consumer Protection Law Group. I’m an AI assistant and I can get you scheduled with our team, or connect you to a person anytime you’d like.” Two seconds. Callers barely react, and the ones who want a human say so and get one.
The harder rule is what the AI must never do: give legal advice. Qualification and scheduling only. “Do I have a case?” gets scheduled, not answered. Write that boundary into the agent’s instructions explicitly, then audit transcripts to confirm it holds. An AI that answers legal questions on your firm’s phone line is practicing law under your name.
The economics
A receptionist costs $3,000 to $4,000 a month and covers 40 hours a week. Voice AI covers 168 hours a week for $200 to $500 a month.
That’s not a 10x improvement on one axis. It’s 4x the coverage at a tenth of the cost, before you count the calls a busy receptionist misses during the hours she does work. Every missed call in a competitive practice area is a lead your competitor’s phone answered.
The receptionist doesn’t disappear from this math. She stops answering “what are your hours” for the ninth time that day and starts working the queue of signed-ready leads the AI qualified overnight. Cheapest close-rate improvement I know of.
What still needs a human
Emotional first calls. Someone whose kid was just injured, or who is calling a family lawyer for the first time, should reach a person. Route those practice areas and those scenarios to humans by rule.
Retainer closes. The AI books the consultation. A human converts it.
Judgment calls. Conflicts questions, press inquiries, anything where the right answer depends on context the script doesn’t have. The AI’s job on those calls is a fast, clean handoff.
How to start without betting the firm
Don’t rip out your phone tree on day one. Forward after-hours calls to a voice AI first. Those calls currently go to voicemail, and most voicemails never convert, so the downside is nearly zero and the comparison is honest: AI versus nothing.
Then do the unglamorous part. Read every transcript for the first two weeks. You’ll find the questions your script doesn’t cover, the moments callers get confused, and the one weird thing people in your practice area always ask. Tune the script weekly. By week four you’ll trust it with overflow during business hours. Most firms never go back.
The checklist: forward after-hours calls first, disclose AI on every call, read every transcript for two weeks, and keep the agent away from legal advice. That sequence gets you the coverage without the risk.
Tonight, after your office closes, call your own firm. Listen to what a potential client hears. If it’s voicemail, you’ve found this week’s project: pick one of the three platform routes above and put an AI on your after-hours line by Friday.